benchr
OpenAI previewed GPT-5.6 on June 26 to a small government-vetted partner group. On July 9, the API changelog moved the family into general availability. The important live-doc changes are straightforward: GPT-5.6 now appears on OpenAI's public models and pricing pages, the gpt-5.6 alias routes to Sol, and all three tiers list 1.05M context with 128K max output.
Three models, and what each is for
GPT-5.6 is a series, not a single model. OpenAI split it three ways by cost and capability:
- Sol — the flagship. This is the model OpenAI is making frontier claims about, and the one the government review is built around.
- Terra — the balanced tier. OpenAI describes it as roughly GPT-5.5-class capability at about half the cost, which is the headline efficiency story of the release.
- Luna — the cheapest and fastest. The high-volume option for work that doesn't need the flagship.
The naming is a deliberate break from OpenAI's "Mini / Nano" suffixes. Sol, Terra, and Luna read as a tiered family rather than shrunk-down variants — a structural echo of how Anthropic ranks Opus, Sonnet, and Haiku.
Two new reasoning modes: "max" and "ultra"
Alongside the three models, GPT-5.6 introduces two reasoning modes that change how a model works rather than which model you pick. Max dials up reasoning effort — deeper, longer deliberation on a single line of work. Ultra goes wider instead of deeper: it spins up subagents to parallelize a complex task across multiple workers, which is the same architectural idea behind OpenAI's recent agentic Codex push.
What OpenAI has not said is what either mode does to your bill or your latency. Deeper reasoning and parallel subagents both tend to burn more output tokens, sometimes far more, but no per-mode pricing or token-multiplier has been published. Treat the cost of "max" and "ultra" as unknown until OpenAI documents it.
The pricing — official on OpenAI's API page, and lower than at launch
OpenAI's pricing page lists GPT-5.6 directly, and cut two of the three tiers on July 30, 2026. Standard short-context rates per million tokens are now Sol $5/$30 (unchanged), Terra $2/$12 (was $2.50/$15), and Luna $0.20/$1.20 (was $1/$6).
| Model | Input | Output | Cached input |
|---|---|---|---|
| Sol (flagship) | $5.00 | $30.00 | $0.50 |
| Terra (balanced) | $2.00 | $12.00 | $0.20 |
| Luna (cheapest) | $0.20 | $1.20 | $0.02 |
Cached input is $0.50, $0.20, and $0.02 for Sol, Terra, and Luna. OpenAI has not restated batch, flex, or priority pricing since the July 30 cut.
The sticker prices are most legible against the rest of the field, so here is where each tier lands now. Sol's $5 / $30 is still the identical sticker to GPT-5.5 ($5 / $30) — that comparison didn't change. Terra's $2 / $12 now undercuts GPT-5.4's $2.50/$15 outright, which makes the "GPT-5.5-class capability" pitch stronger than it was at launch. Luna's $0.20 / $1.20 now beats GPT-5 Mini ($0.25 / $2) on both sides, having started out between GPT-5 ($1.25 / $10) and Mini. For external anchors: Claude Opus 4.8 is $5 / $25, Gemini 3.1 Pro is $2 / $12, and DeepSeek V4-Pro is $0.435 / $0.87 — Luna's new input rate beats DeepSeek, though DeepSeek still wins on output. The full breakdown, with the caching math and these caveats restated, lives on the GPT-5.6 pricing page.
Who can use it: API customers
The June preview was limited to roughly 20 trusted partners. That changed with the July 9 API release: GPT-5.6 is now listed in OpenAI's public API docs. Existing customers still need to check their account, region, and organization settings, but the model family is no longer described as preview-only.
The government-gated preview remains part of the release history, and it explains why benchr originally treated availability cautiously. The live source of truth is now OpenAI's API changelog, models page, and pricing page.
The benchmarks OpenAI published
At the preview, the numbers were a black box — but OpenAI's GPT-5.6 preview system card publishes the scorecards, and they back the headline claim. On Terminal-Bench 2.1, the agentic command-line benchmark, Sol sets a new state of the art — and in ultra mode it pulls clear of the field. Here is the full ranking, reproduced from OpenAI's own charts.
TerminalBench 2.1
Agentic command line · higher is betterSource: OpenAI GPT-5.6 preview system card. Purple = GPT-5.6 family; Sol's ultra mode runs subagents in parallel.
The other two benchmarks are efficiency frontiers — they plot score against the output tokens spent, so a curve that sits higher and further left is doing more with less. On both, Sol leads, and the three GPT-5.6 tiers fan out by how much reasoning each is willing to spend.
GeneBench v1
Biology · score vs output tokensPeaks: Sol ~30.7% · Terra ~28.3% · GPT-5.5 ~23% · Luna ~14.5%. Curves reproduced from OpenAI's preview system card; the point is the shape — Sol reaches the top scores spending the fewest output tokens.
ExploitGym
Cyber · intended exploits vs output tokens (6h limit)Peaks (6-hour limit): Sol ~33.7% · Terra ~23.3% · GPT-5.5 ~15.2% · Luna ~12.4% · GPT-5.4 ~7%. Dashed = the 6-hour-budget frontier; reproduced from OpenAI's preview system card.
So the SOTA claim holds up: Sol leads Terminal-Bench 2.1 outright, and ultra mode stretches the lead to 91.9%. Read the field carefully, though — Terra lands at 84.3%, exactly tying Claude Fable 5 and barely ahead of GPT-5.5's 83.4%, while Luna's 82.5% slots just below GPT-5.5. The series wins at the top; in the middle it's a crowded few points. These Terminal-Bench 2.1 figures are unchanged from the June 26 preview.
SWE-bench Verified and GPQA Diamond, added to the system card
OpenAI's preview system card also carries two scorecards that weren't there when benchr first covered this preview: SWE-bench Verified and GPQA Diamond, for all three tiers.
| Model | SWE-bench Verified | GPQA Diamond |
|---|---|---|
| GPT-5.6 Sol | 89.8% | 91.2% |
| GPT-5.6 Terra | 85.2% | 88.0% |
| GPT-5.6 Luna | 79.8% | 82.0% |
Sol's Terminal-Bench 2.1 lead doesn't carry over to either new benchmark: its 89.8% on SWE-bench Verified trails Claude Fable 5's 91.0%, and its 91.2% on GPQA Diamond trails Claude Opus 4.8's 93.6%. The same system card lists Sol's broader eval sweep: LMSYS Arena 1440, MMLU 93.8%, HumanEval 95.5%, MATH 96.0%, and ARC-AGI-2 23.5.
OpenAI's live models page now confirms the context window — 1,050,000 tokens for Sol, Terra, and Luna — the maximum output length — 128,000 tokens for all three tiers — and the exact API model IDs: gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna. GPT-5.6 Sol remains benchr's canonical models.json entry for the series; Terra and Luna stay in model-figures.json.
What it means for you
The practical answer has changed: GPT-5.6 is now callable through the API where your account has access. Sol leads Terminal-Bench 2.1 (88.8%, 91.9% in ultra mode) but trails Claude Fable 5 on SWE-bench Verified and Claude Opus 4.8 on GPQA Diamond, so the model choice is still workload-specific rather than automatic.
If you're an existing OpenAI shop, Terra is the tier worth testing first: it now undercuts the old GPT-5.4 price while keeping SWE-bench Verified 85.2% and GPQA Diamond 88.0%. Luna is the volume tier, and its July 30 price cut to $0.20/$1.20 makes it worth a second look even if you dismissed it at launch.